<b>МЕТОДИКА ОЦЕНКИ СТРУКТУРНОГО РИСКА СВАРНЫХ СОЕДИНЕНИЙ РАМ КУЛЬТИВАТОРОВ-ГЛУБОКОРЫХЛИТЕЛЕЙ В УСЛОВИЯХ ЭКСПЛУАТАЦИИ</b>
The article presents a methodology for assessing the structural risk of welded joints in the frames of deep tillage cultivators under real operating conditions. The research object is the frame of the KPG-3-5 cultivator–subsoiler subjected to static, dynamic, and vibration loads. Traditional inspection methods, including visual examination and selective ultrasonic testing, depend substantially on operator qualification and have limited applicability under field monitoring conditions. To analyze the condition of the joints, a dataset of 412 digital images was formed under field conditions without preliminary surface cleaning. The images were acquired under natural lighting and from different observation angles. This approach preserved operational contamination, corrosion traces, and microdamage characteristic of machinery used under real conditions. Image analysis was performed using a computer vision approach based on a pretrained ResNet-18 neural network and the Grad-CAM visualization method. An integrated structural risk index SR(x) was introduced, combining the geometric characteristics of the weld seam with the activity of model attention zones. The results showed increased SR(x) values in working-shank mounting areas, where the mean value reached 0.71 ± 0.04. Moderate values were obtained for main-beam joints, while secondary structural elements demonstrated low values of 0.32 ± 0.05. The regions of maximum model activation coincided with structurally loaded frame zones exposed to cyclic and impact loads. The proposed methodology enables potentially vulnerable areas to be identified before visually pronounced defects appear and supports the transition from binary assessment to quantitative engineering diagnostics. Its practical significance lies in the possibility of integration into agricultural machinery condition-monitoring systems to support preventive maintenance and reduce unplanned downtime during field operations. Further research will focus on expanding the dataset, statistically calibrating the SR(x) index parameters, and adapting the methodology to other types of agricultural machinery frame structures.
Сведения
- Журнал
- Applied Agro Engineering
- Опубликовано
- 2026 · Т. 1, № 3, С. 40-53
- Язык
- Русский
- ISSN
- 3134-853X · 3134-8548
- Издатель
- West Kazakhstan Agrarian Technical University named after Zhangir Khan